Our sampling site is located on Mt. Lama (119.53°E, 42.19°N), which is a rocky hill with shallow soils. Very few Chinese pines (Pinus tabulaeformis), a species widely employed in dendroclimatological studies in northern China, are distributed sporadically on Mt. Lama. A total of 32 cores were collected from 17 healthy Chinese pines by increment corers in the summer of 2013. According to standard dendrochronological methods, these cores were fixed and then surfaced by sandpaper until the yearly boundaries were easily to distinguish under a microscope (Stokes and Smiley, 1996). Tree-ring-width of each core was measured by using the LINTAB 5 with a resolution of 0.01 mm and crossdated by using TSAP software. We used the COFECHA program to control the quality of crossdating and ring-width measurements (Holmes, 1983).

The ARSTAN program was employed to establish the ring-width chronologies (Cook and Kairiukstis, 1990). The traditional linear and negative exponential function method was selected to remove the non-climate signals. To retain both low and high frequency climate informations, the STD chronology was chosen for climate response analysis and climate reconstruction. The reliable period of the chronology starts from 1710, after which the subsample signal strength (SSS) is greater than 0.85 (Wigley et al., 1984).

The climatic data from four meteorological stations, Chifeng (118.5°E, 42.18°N, 668.6 m a.s.l.), Chaoyang (120.26°N, 41.33°E, 174.3 m a.s.l.), Yebaishou (119.37°N, 41.25°E, 422 m a.s.l.) and Baoguotu (120.42°N, 42.2°E, 400.5 m a.s.l), were used to generate a regional climate dataset (Figures 1 and 3). The data from the four stations show similar variation patterns characterized by rainy and hot summers and dry and cold winters. The average climate data from the four stations show that the annual precipitation is 360 mm, 69% of which is concentrated in summer. These are typical climate characteristics in the marginal region of the EASM. The annual average temperature of the study region increased by approximately 1.5°C, and the rainfall decreased by approximately 85 mm from 1960 to 2015 (Wang et al., 2015), implying warmer and drier climate conditions in recent years.

The PDSI was designed to represent drought conditions by using a water balance model which takes account of water supply (precipitation), demand (evapotranspiration) and loss (runoff) (Palmer, 1965; Dai et al., 2004). The PDSI records usually show strong association with tree-ring chronologies, especially in arid and semiarid regions. The nearby self-calibrating PDSI (scPDSI) grid data (43.75°N, 118.75°E) (Dai et al., 2011) since 1946 AD, before which there were many missing values, were used in this study.

Pearson correlation analysis was performed to examine the relationship between the STD chronology and climatic data. The stability and reliability of the reconstruction was validated by the split calibration-verification method (Meko and Graybill, 1995). Various statistical parameters, for example, sign test (ST), reduction of error (RE) and coefficient of efficiency (CE), were employed to evaluate the transfer function performance. Positive values of the RE and CE usually denote acceptable model skill (Cook et al., 1999). A spatial correlation analysis was conducted to evaluate the regional representativeness of Lama PDSI reconstruction using the University Corporation for Atmospheric Research (UCAR) scPDSI dataset (http://climexp.knmi.nl).

The calculations indicate that the Lama STD chronology is negatively correlated with temperature and positively correlated with precipitation in most months. This climate-growth pattern is typical in semiarid regions of China and has been found in a number of dendroclimatological studies (Fang et al., 2010; Song et al., 2011; Mei et al., 2019). Significant correlations between the STD chronology and climatic factors, e.g., precipitation and temperature, mainly concentrated during May-July, which suggests that the hydrothermal conditions of the fast-growing season play important roles in radial growth (Figure 4a). After different combinations, the Lama STD chronology showed the highest correlation with the total precipitation from prior August to current July, r=0.593 (1953-2012, p<0.001), whereas May-June temperature showed the highest correlation with tree-ring series r=-0.549 (1953-2012, p<0.001) (Figure 4a).

The monthly PDSI and Lama STD chronology are highly correlated from prior August to current September (Figure 4b). PDSI series are calculated based on observed precipitation and temperature records, and thus it can represent soil moisture (Palmer, 1965). During the growing season in our study region, drought events are usually caused by low precipitation and high temperature, which can further increase soil water evaporation. Severe water stress can lead to tree growth decline and even mortality. After comparing various combinations of different climatic factors, the Lama STD chronology showed the highest correlation with May-August mean PDSI r=0.695 (1946-2012, p<0.001).

The local May-August PDSI was reconstructed by using the STD chronology from Mt. Lama. The transfer function is as follows:

PDSI₅₈=3.926W_t-4.9056

(n=67; r=0.695; R²=48.0%; R²_adj=47.2%; D/W=1.091; p<0.001).

PDSI₅₈ is the mean PDSI during May-August, W_t represents the Lama STD chronology in year t. The reconstruction function explains 47.2% of the actual PDSI changes between 1946 and 2012. The statistics of split calibration-verification analysis (Table 1), e.g., positive values of RE and CE, demonstrate the reliability and stability of the transfer function over time. Therefore, the May to August PDSI for Mt. Lama was reconstructed from 1710 to 2012 (Figure 5). The reconstructed PDSI₅₈ shows similar variation to the actual PDSI records at high and low frequencies (Figure 5).

Our PDSI reconstruction displays the drought variations on Mt. Lama at the annual to decadal scales. To assess the occurrence of drought events, we defined drought years as PDSI values below -2.50 (mean-1σ), and wet years as PDSI values higher than 0.43 (mean+1σ). There have been nine dry years since the 1950s, which accounted for 18.4% of the total 49 dry years, indicating an increase in drought intensity during recent years. The ten most severe drought events occurred in 1861, 1867, 1724, 1832, 1885, 1748, 1778, 1968, 1900, and 1723. The worst drought in 1861 has rarely been reported in previous studies but was found in a nearby tree-ring record (Liu et al., 2011) and regional historical documents, such as in Inner Mongolia, Hebei and Shanxi (Wen, 2008), implying that it may be a regional drought event. The late 1920s drought, the worst drought in northwestern and northern China (Liang et al., 2006; Liu et al., 2020), is not prominent on Mt. Lama. The 1900 drought is widely discovered in tree-ring records in northern China (Liu et al., 2017; Mei et al., 2019) but is not found in northwestern China (Liu et al., 2013). All these comparisons demonstrate the spatial heterogeneity of annual hydroclimate variability.

On the decadal scale, the reconstructed drought series displays an obvious drying tendency since the 1950s, which has been found in a number of tree-ring hydroclimate reconstructions in northern China (Li et al., 2006; Liu et al., 2020). The increase in anthropogenic sulfate aerosols has been proven to be related to the drying trend in the northwestern marginal area of the EASM (Liu et al., 2019). The Yellow River, the longest in northern China, also showed an unprecedented decreasing since the 1970s, which was attributed to the decreasing of the Asian monsoon precipitation and increasing of anthropogenic water consumption especially the agricultural irrigation (Liu et al., 2020). Summer heatwave and soil moisture reconstruction over inner East Asia showed that the severe drought conditions would continue in the future (Zhang et al., 2019), which should be paid more attention by scientists and governments. The most severe drought of the reconstructed series occurred during the 1860s with a duration of 8 years, which has been identified by other tree-ring series in the northern marginal region of EASM, such as in Chifeng-Weichang (Liu et al., 2010), Kalaqin (Liu et al., 2011) and Lamadong in Inner Mongolia (Liu et al., 2017). The association among them reflects the synchronous hydroclimate variability over the northern marginal region of the EASM.

The annual precipitation variability has been reconstructed based on STD chronology at Kalaqin (Liu et al., 2011) which is approximately 100 km from Mt. Lama. The two hydroclimate records show good similarity at annual to decadal scales (Figure 6). The annual correlation between them was significant, with r=0.68 (n=238, p<0.001). This comparison not only verified our drought reconstruction, but also reflected coherent hydroclimatic variability, which was attributed to the influence of the EASM. Consistent spatial correlation patterns (Figure 7) were revealed by the grid data and the PDSI reconstruction, indicating the Lama PDSI reconstruction can represent the drought variability on the northernmost margin of the EASM. Spatial correlation suggested no obvious relationships between Lama drought reconstructions and the western margin of the EASM, implying different hydroclimatic factors on the northern and western margins of the EASM.

The observed climatic data in our study region show that June-August precipitation ac counts for 69% of yearly precipitation, which is strongly related to EASM. A significant correlation was found between our drought reconstruction and July-September EASM index (Zhao et al., 2015), with r=0.46 (n=65, p<0.001) (Figure 8). However, some inconsistencies were also found, which means that the existing EASM index cannot fully capture the annual hydroclimate variability in the northern marginal region of the EASM. The EASM usually breaks out in the central Indochina Peninsula in late April or early May, then gradually advances northward, and finally arrives at the northernmost position, our study region, in July (Wang and Lin, 2002). The strength of the EASM in different stages is not consistent. Most EASM indexes reflect precipitation variability in southern China, especially the Mei-yu-Changma-Bai-u area (Zhao et al., 2015), and are not suitable to represent the monsoon strength in the marginal region of the EASM. More effort should be made to reconstruct EASM indexes specifically aimed at the EASM marginal region, which is frequently affected by droughts.

A significant positive correlation was shown between our PDSI reconstruction for Mt. Lama and the Asian polar vortex area index during the period of the prior November to the current January, with r=0.50 (n=61, p<0.001) (Figure 8). The comparison (Figure 8) indicates that the large Asian polar vortex area in winter may be related to humid summers in the northern marginal region of the EASM, and vice versa. A previous study also found that the winter polar vortex strongly influenced the precipitation variability over northern China in summer (Huang et al., 2004). The polar vortex is a planetary-scale cyclone around both poles in the middle or high latitudes (Zhu et al., 2018), and usually shows close relationship with the Arctic Oscillation (AO), particularly during winter (Deng et al., 2006; Wu et al., 2008). The winter AO/North Arctic Oscillation (NAO)-forced wave train propagates eastward and strongly influences precipitation over East Asia in summer (Sung et al., 2006), which is a kind of Eurasian teleconnection pattern (EU). The negative EU pattern is accompanied with a blocking event and strong polar vortex which will lead to a strong western Pacific subtropical high and low summer precipitation in northern China (Zhao et al., 2016). Studies have revealed a weakening trend of polar vortex caused by the melting of arctic sea ice in recent decades (Zhang et al., 2016), which will exert strong influence on the precipitation variability in China.